GB2374243A

GB2374243A - An MPLS network with traffic engineering tunnels where the level of traffic is monitored and the tunnel bandwidth adjusted in response

Info

Publication number: GB2374243A
Application number: GB0128825A
Authority: GB
Inventors: Andrew Reeve
Original assignee: Roke Manor Research Ltd
Current assignee: Roke Manor Research Ltd
Priority date: 2001-04-04
Filing date: 2001-12-03
Publication date: 2002-10-09
Anticipated expiration: 2021-12-03
Also published as: GB2374243B; GB0128825D0; GB0108461D0

Abstract

An automated network comprises a plurality of boundary nodes connected to at least one other network node, each boundary node being connected to at least one other boundary node via a traffic engineering tunnel, the tunnel comprising a source boundary node and a destination boundary node. Each source boundary node monitors traffic flow along each tunnel to which it is connected, and a tunnel parameter is modified in response to a change in traffic flow. The parameter which is adjusted is typically tunnel bandwidth. The system typically operates in a Multiprotocol label switched (MPLS) environment.

Description

AUTOMATED NETWORK AND METHOD OF DATA ROUTING This invention relates to an automated network and a method of data routing, particularly for mobile phone networks.

Next generation packet networks need to deliver Quality of Service (QoS) to subscriber data according to the needs of particular applications or subscription to premium services. It is desirable that they should support Class of Service (COS), providing relative treatment according to subscription level and adhere to Service Level Agreements (SLAs) which have been created with subscribers and other networks maintain efficient use of expensive resources (nodes and link capacity). Prior art methods such as shortest path routing become congested because all data tries to use the shortest path. One technique to overcome this problem is through the use of multiprotocol label switched (MPLS) traffic engineering (TE) tunnels. These force data to follow specified tunnels between start and end points, not necessarily the shortest path.

As long as any constraints are met e. g. maximum delay, then data is told where to go.

MPLS traffic engineering is a powerful technique and is an essential building block in the provision of QoS and achieving network efficiency. However operations staff will need to monitor network performance and provision of TE tunnels as required. This must be done in the presence of changing and unpredictable traffic patterns, making it a difficult and potentially error prone task.

In accordance with a first aspect of the present invention, an automated network, comprises a plurality of boundary nodes connected to at least one other network node, each boundary node being connected to at least one other boundary node via a traffic engineering tunnel, the tunnel comprising a source boundary node and a destination boundary node; wherein each source boundary node monitors traffic flow along each tunnel to which it is connected, and wherein a tunnel parameter is modified in response to a change in traffic flow.

In accordance with a second aspect of the present invention, a method of automated data routing comprises identifying boundary nodes in a network; establishing connection tunnels between a source boundary node and a destination boundary node, the tunnels having at least one predetermined parameter; wherein in use, traffic flow through the tunnels is monitored and the or each tunnel parameter is modified in response to a change in demand.

Preferably, the tunnel parameter is the bandwidth available on the tunnel, although other parameters such as scheduling priority could be modified.

The source boundary node may monitor and initiate changes in the tunnel parameter, but preferably, the network further comprises a network router to analyse the traffic flow monitored by the or each source boundary node and initiate changes in tunnel parameter.

Preferably, the tunnels are established using MPLS, although other routing systems such as ATM PNNI, RSVP, QoS Routing or COPS could be used.

Preferably, hysteresis is applied to the traffic flow monitoring, such that tunnel reconfiguration activity is dampened.

Preferably, tunnels are allocated maximum and minimum bandwidth limits.

An example of an automated network and a method of data routing according to the present invention will now be described with reference to the accompanying drawings in which: Fig 1 illustrates an example of conventional shortest path routing; Fig 2 illustrates use of traffic engineering tunnels in a network; Fig. 3 illustrates a network operated by the method of the present invention; and, Fig. 4 shows an example of an optical network operated in accordance with the method of the present invention.

Conventionally, an internet protocol (IP) network will employ a shortest path routing protocol such as open shortest path first (OSPF). This can often result in inefficient use of the network resources. In Fig. 1 nodes A-C are core routers and L-R are edge routers with similar capacity and subscriber traffic load. The links A-B, B-C and C-A all have the same capacity. Consider router A. Use of a shortest path routing protocol means that at A all packets from P, Q and R, destined to M, N or 0 will be routed over link A-B, as will packets from M, N and 0 to P, Q or R. Thus links A-C and B-C will only carry packets to or from L. This means that links A-C and B-C have wasted capacity while link A-B is liable to congestion. The Shortest Path problem can cause considerable inefficiency in real networks.

Fig. 2 depicts a network with the same topology as that in Fig. 1 employing MPLS in the core. A, B and C are MPLS switches/routers. MPLS supports the creation of so-called traffic engineering (TE) tunnels to carry traffic between nodes. Calculation of the routes taken by the tunnels ensures that QoS and network performance objectives

are met. For example in Fig. 2 tunnel R-A-B-O could be set up for traffic from R to 0, whereas for traffic between P and M the tunnel P-A-C-B-M, is set up, disjoint from R A-B-O, thus avoiding congestion on link A-B, and ensuring that capacity on links A-C and C-B is used.

MPLS traffic engineering is currently being standardised in the Internet Engineering Task Force (IETF) where the necessary routing protocols and signalling protocols are being defined. It will allow simplified tunnel creation. Tunnel creation is initiated at the source node and requires the specification of the following information, tunnel endpoints (source and destination), bandwidth required, QoS required. The network will calculate a suitable route for the tunnel and perform the necessary signalling to set it up. A tag is put onto a packet with the tunnel identity (id) and each router sends the packet the correct way without any need to worry about IP address. In creating tunnels, sufficient bandwidth must be reserved and planners have to put a lot of effort in to deciding where to send traffic.

The route taken by a tunnel is calculated using an algorithm such as Constrained Shortest Path First (CSPF). By applying appropriate constraints during route calculation, CSPF can ensure that the traffic carried in a tunnel will receive the required QoS and that network efficiency goals are achieved. In particular, the routes taken by traffic engineering tunnels can be calculated such that congestion is avoided on the physical links.

However, it is necessary to take into account a number of factors, either alone or in combination. These include dramatic changes in traffic patterns caused by sudden events (e. g. publication of the Starr Report, release of a new game or movie); hourly, daily, weekly and seasonal changes in traffic patterns; growth in subscribers, addition of new servers, network growth and change; emergence of new applications and modes of work; server and network failures. These multiple factors can overwhelm the human operators and tend to lead to compromise routing decisions being made. However, the present invention addresses the problems of prior art systems without the need for operator intervention.

The first stage is to identify which of the nodes are traffic engineering region boundary nodes. A boundary node is where the tunnel begins. Outside the boundary is a standard routed network. Boundary nodes include nodes which have interfaces to subscribers, access network servers or external networks. In the example shown in Fig.

3, the boundary nodes are A, B & C. The next step is to identify what TE tunnels are

required to connect the boundary nodes. By default the boundary nodes would be fully meshed, but connectivity could be reduced by application of network policy. One logical route could have different physical routes for different uses, e. g. voice or internet. For a large network with lots of nodes, different types of traffic may be grouped together. A joint tunnel is allocated if there are only small amounts of traffic of that type and separate tunnels if there is a lot of traffic of that type. For a small network, a bandwidth broker links to each of the boundary nodes, records statistics and sets tunnel bandwidth for the whole network. This works well for a small network, but is not scaleable for a large one. Also there is a single point of failure, so there needs to be a shadow system. Alternatively, each router looks at the network and makes its own independent decisions, but there would need to be bandwidth quotas for sending in and receiving.

Traffic statistics on how often, how much and what type of traffic passes are recorded for each tunnel that has been defined. This record is nearly instantaneous and data is only kept for long enough to make a decision about whether a change is required. Data is not intended to form a historical record. From the statistics, bandwidth requirements for each tunnel can be calculated. For example, real time traffic needs more available bandwidth than it will actually use, to prevent loss of data due to congestion, whereas bandwidth for web browsing can be allocated with far smaller tolerances. Preferably, a network wide policy is applied to calculate the bandwidth requirements or allowances for different services and traffic types. Standard MPLS with a constrained SPF algorithm may be used. The constraints are entered and MPLS decides on the route. The statistics are continuously monitored and the bandwidth requirement changed if necessary. Periodic recalculation is performed as a matter of course. However, if sudden dramatic changes are detected, through the use of thresholds (e. g. queue fill thresholds), this would trigger an immediate recalculation for affected tunnels and the tunnels are modified accordingly, using MPLS services.

Tunnel reconfiguration activity is controlled by applying hysteresis. Additional data relating to statistics and state information (routes and bandwidths) may be recorded to allow monitoring and control of the network by an operator, for example to allow the operator to monitor and change to automation gradually, or to assess where problems were in their system. The method of the present invention enables an automated network to operate in which minimum and maximum bandwidths may be specified for the tunnels and thereafter actual bandwidths are determined by a source boundary node.

Such maxima and minima may be specified globally or on a per tunnel basis. By this means the operator can specify certain bandwidth allocations, but devolve use of the remaining capacity to the automated network. This is particularly important following initial deployment of the present invention, when an operator may be reluctant to hand over full control to an automated system.

The present invention allows bandwidth adjustments to be achieved in seconds, rather than in hours or days as required by traditional techniques needing operator intervention. This means that following dramatic changes in traffic patterns, the minimum disruption occurs and users get a faster service more quickly. The automated network and routing method ensures the most efficient use of all resources is made, instead of the compromises inherent in any operator-controlled system. There is also a dramatic reduction in the cost of ownership, since specialist engineers are not required to calculate traffic requirements and plan tunnels. Also, fewer operators are required to manage the network and perform configuration tasks. Overall network efficiency is increased and full use is made of the deployed infrastructure (i. e. expensive fibre, core routers and switches).

Since MPLS is the basis for traffic engineering in packet networks and for connection signalling in future optical networks, the automated network of the present invention provides a means for integrated and cost effective bandwidth management of core networks comprising both high capacity routers and optical switches. MPLS supports tunnel protection and restoration, so resilience to failure and bandwidth management can be provided in a fully integrated fashion. Network upgrades are simplified, because once new equipment and links are installed, the network adjusts automatically.

An important feature of the invention is that the QoS of the most important traffic is maintained during failure and traffic peaks. For small networks centralised control by a traffic engineering controller (TEC) is feasible. The TEC would monitor the traffic patterns by periodic upload of statistics from the TBNs. Periodically, or following sudden dramatic change, it would recalculate the ideal configuration of TE tunnels. Overall network policy and capacity quotas are implemented at this stage. The TEC implements changes by issuing tunnel management commands to the TBNs.

TBNs maintain FEC statistics and periodically re-calculate tunnel requirements. Each TBN create, modify or delete TE tunnels as required. TBNs would have transmit and

receive quotas which they must adhere to when initiating and accepting tunnels respectively.

Two basic techniques for calculating the bandwidth requirements for a tunnel are described below. In one example, the router calculates the absolute bandwidth requirement based on the traffic load. When maintaining statistics, traffic is sorted into buckets for counting purposes. With three buckets split into real-time (R), upper class (U) and best effort (B), then weighting factors (a, P, y) can be applied as follows.

bandwidth = aR + PU + YB Other formulae and techniques for calculating the bandwidth could be used.

The calculated bandwidth would normally be rounded up to a suitable quantised value and the new or modified tunnel would be requested accordingly.

An alternative is a quota-based system for calculating bandwidths. Each router is allocated a bandwidth quota for the sum of all traffic to and from all of its peers.

The absolute traffic requirement is calculated for each tunnel using a method such as that described above. Tunnel bandwidths are calculated by application of a fair-share algorithm which takes into account the tunnel requirements, but ensures that no router's quota is exceeded for either traffic transmitted or received. This would be an iterative process. An algorithm such as that described in British Patent publication no. 2362778 could be employed.

The method of the present invention is applicable to both packet networks and optical networks. An example of an optical network is shown in Fig. 4. It shows an IP network with an optical core consisting of optical cross-connects and DWDM links.

The DWDM links carry a number of high bandwidth channels (wavelengths), for example 32 10 Gigabit/second wavelengths. The routers are inter-connected by light paths or optical connections, each with, say 10 Gigabit/second capacity. A number of such optical connections may be set up between each pair of routers. With the emergence of the Automatically Switched Optical Network (ASON) the creation of such connections will be automated and individual routers will be able to initiate connections in the order of a second.

The optical connections perform a similar function to MPLS tunnels. In fact the set of optical connections between a pair of routers may be considered as a bundle whose capacity is determined by the number of connections (wavelengths).

Furthermore, there is a requirement to determine how to most effectively distribute the

available bandwidth (wavelengths) between the router interconnects, and that this must depend on the particular traffic requirements at the time. The method of the present invention is an effective way to perform this bandwidth management. In this case, the routers monitor their traffic load, calculate their optical bandwidth requirements, and create and tear down optical connections accordingly.

MPLS, which originated in the packet world, is being enhanced (so-called Generalised MPLS) to support the signalling requirements of optical networks and indeed ASON, will be based on GMPLS.

Claims

CLAIMS 1. An automated network, the network comprising a plurality of boundary nodes connected to at least one other network node, each boundary node being connected to at least one other boundary node via a traffic engineering tunnel, the tunnel comprising a source boundary node and a destination boundary node; wherein each source boundary node monitors traffic flow along each tunnel to which it is connected, and wherein a tunnel parameter is modified in response to a change in traffic flow.
2. An automated network according to claim 1, wherein the tunnel parameter is the bandwidth available on the tunnel.
3. An automated network according to claim 1 or claim 2, further comprising a network router to analyse the traffic flow monitored by the or each source boundary node and initiate changes in tunnel parameter.
4. A method of automated data routing, the method comprising identifying boundary nodes in a network; establishing connection tunnels between a source boundary node and a destination boundary node, the tunnels having at least one predetermined parameter; wherein in use, traffic flow through the tunnels is monitored and the or each tunnel parameter is modified in response to a change in demand.
5. A method according to claim 4, wherein the at least one tunnel parameter comprises bandwidth.
6. A method according to claim 4 or claim 5, wherein the tunnels are established using MPLS.
7. A method according to any preceding claim, wherein hysteresis is applied to the traffic flow monitoring, such that tunnel reconfiguration activity is dampened.
8. A method according to any preceding claim, wherein tunnels are allocated maximum and minimum bandwidth limits.
9. A method according to any of claims 4 to 8, wherein a router is provided to analyse the traffic flow monitored by the source boundary node and initiate changes to the tunnel parameter.
10. A method according to claim 5, wherein an absolute bandwidth requirement is calculated based on traffic load.
11. A method according to claim 10, wherein the bandwidth available on the tunnel is calculated by application of a fair share algorithm to the absolute bandwidth requirement.
12. A method according to claim 10 or claim 11, wherein the absolute bandwidth requirement is equal to the sum of each traffic type multiplied by their respective weighting factors.
13. A method according to claim 12, wherein the traffic types comprise real time, upper class and best efforts.